Steel defects classification based on vision attention

Steel surface defects are imperfections or irregularities that arise during the production, processing, or finishing of steel products. Surface defects degrade the robustness and stability of steel, compromise its structural integrity. As steel is a widely -used material across the industries, the quality assurance of steel is extremely important. Traditional defect detection methods are prone to human errors and difficult to detect small or subtle defects. Deep learning offers a promising solution in automated defect detection and classification, especially Vision Transformer. This paper implements different pre-trained deep learning models to classify steel surface defects from NEU dataset. The Vision Transformer has the best performance with the highest classification accuracy of 99.17% and the least computation time of 229s. Vision Transformer successfully capture the complex patterns in steel defects and being able to classify them accurately with the powerful attention mechanism.